Bug Report & Feature Request: Fix Gemini Search Conflict + Add Code Execution & URL Context

The built-in Web Search tool by model providers is great news and great start for future updates!

However, there is currently a problem when using the Gemini API: enabling Model Tools (Grounding) and Plugins simultaneously throws a400error: "Gemini API does not support using Grounding Search and plugins together at the same time. Please turn off one of them and try again."

1. The Fix: Switch to "Tool-Based" Search

   I believe this error occurs because the app currently uses the legacy google_search_retrieval configuration parameter. According to the Gemini documentation, this legacy method is mutually exclusive with function_declarations (Plugins).
   
   To fix this for Gemini 2.0 and later models, please switch to the modern Tool-based implementation:

   • Instead of a separate retrieval config, define Google Search as a standard item inside the tools list.

   • This allows Google Search to sit alongside function_declarations in the same request, resolving the conflict and allowing the model to dynamically choose between searching or using a plugin.

2. Feature Request: Code Execution & URL Context

   Since this update involves modifying the tools definitions, could you also add other native Gemini tools?

   • Code Execution: This allows the model to generate and run Python code iteratively for complex math and logic.

   • URL Context: This is highly synergistic with Search. As noted in the Google documentation: "When both URL context and Grounding with Google Search are enabled, the model can use its search capabilities to find relevant information online and then use the URL context tool to get a more in-depth understanding of the pages it finds."

Thank you for your continuous improvements to TypingMind!